02.03.2013 Views

Target Discovery and Validation Reviews and Protocols

Target Discovery and Validation Reviews and Protocols

Target Discovery and Validation Reviews and Protocols

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

52 Imoto et al.<br />

Table 3<br />

List of Root Genes of Identified Drug-Active Pathways (13)<br />

Gene name GO annotations<br />

CHA1 Serine family amino acid catabolism, mitochondrial nucleoid<br />

CAF16 Regulation of transcription, cytoplasm<br />

PNC1 Chromatin silencing, cytoplasm<br />

ADE5,7 Purine base metabolism, cytoplasm<br />

PIL1 Response to heat, cytoplasm<br />

KGD1 Tricarboxylic acid cycle, mitochondrial matrix<br />

FRE1 Iron ion transport, plasma membrane<br />

ADE13 Purine base metabolism<br />

IZH2 Lipid metabolism, integral to membrane<br />

GO, gene ontology.<br />

relations correctly, when the data have sufficient accuracy <strong>and</strong> information. However,<br />

this method requires discretization of the expression values into two levels (0 or 1),<br />

<strong>and</strong> the quantization probably causes information loss. By choosing the threshold η<br />

appropriately in a heuristic manner, it is reported that this naive algorithm works well<br />

in practice, although the method is not theoretically well founded.<br />

3. In practice, biological experiments that disrupt the target druggable genes are<br />

needed for confirming the results of the analysis.<br />

4. For identifying drug-affected genes by the virtual gene technique, we simply consider<br />

genes that have five or more virtual genes as the parent genes as the putative<br />

drug-affected genes. That is, genes are under direct influence of the virtual genes.<br />

However, a gene that has only one virtual gene as its parent may be the primary<br />

drug-affected gene, depending on the mode of action for a given drug, <strong>and</strong> this<br />

possibility must be analyzed case by case.<br />

5. In Bayesian network literature, for example, Neapolitan (25), it is shown that<br />

determining the optimal network is an NP-hard problem. However, recent development<br />

of supercomputer technology enables us to optimize the small gene network,<br />

including 40 or fewer genes, by using a suitable algorithm (26). However, for<br />

larger numbers of genes, we use a heuristic strategy such as a greedy hill-climbing<br />

algorithm to discipher graph structure.<br />

6. The simplest way for discretizing microarray gene expression data is the use of<br />

some thresholds: let {u 1 ,…,u m } be a finite set of discrete values <strong>and</strong> {I 1 ,…, I m }<br />

be regions satisfying j I j = R <strong>and</strong> I i I j =∅for i ≠ j. An expression value x ij is<br />

then transformed to u j when x ij ∈ I j (see Friedman et al. [6]). An alternative way<br />

for discretization is the use of k-means clustering (8). Also, Shmulevich <strong>and</strong> Zhang<br />

(27) proposed a method for binalizing gene expression data.<br />

7. The nonparametric regression model in the Bayesian networks of Imoto et al. (9,10)<br />

is an additive regression model. However, in reality, there are several known<br />

transcriptional relationships that are more additive. Imoto et al. (28) proposed a

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!